Anthropic’s Claude Opus 5 set a new ARC-AGI-3 high score, raising fresh questions about real reasoning gains versus benchmark targeting.

OpenAI says about 30% of SWE-Bench Pro tasks may be broken, raising new doubts about how the AI industry measures coding models.

A new generative AI-powered benchmarking system reveals China's dominance in the autonomous vehicle and robotaxi sector over US competitors.

Google DeepMind launches Werewolf and poker benchmarks on Kaggle Game Arena to test AI social skills, deception detection, and risk management. Gemini 3 Pro and Flash models demonstrate significant performance leap over previous generation.